Forecast backtest (MAPE + RMSE)

Backtest a forecasting method on the input series by holding out the last `testSize` observations, forecasting them, and computing MAPE (mean absolute percentage error) + RMSE (root mean squared error). Send POST /api/forecast-eval with the required fields values, testSize and method and pay $0.001 per call over x402 or MPP, or call it free by solving a proof-of-work challenge. It returns a JSON object with method, n, testSize, trainSize, mape and 3 more.

Lets an agent pick which method (mean / naive / drift / ses / holt / holt-winters) actually fits its data before committing to a forward forecast. Always returns a `warnings` array - empty when the backtest is well-posed, populated when `testSize` exceeds n/2 (treat error as indicative not predictive).

Category: Live public data · Tags: forecast backtest evaluation mape rmse cross-validation

TRY IN PLAYGROUND →

Parameters

Example request

curl -i -X POST https://agent402.tools/api/forecast-eval \
  -H "Content-Type: application/json" \
  -d '{"values":[10,12,13,12,15,16,18,19,21,22],"testSize":3,"method":"drift"}'

Without payment this returns HTTP 402 Payment Required with the exact price for forecast-eval; any x402 v2 or MPP client pays it and retries.

Example response

{
  "method": "drift",
  "n": 10,
  "testSize": 3,
  "trainSize": 7,
  "mape": 1.1139,
  "rmse": 0.2722,
  "forecast": [
    {
      "step": 1,
      "actual": 19,
      "predicted": 19.3333
    },
    {
      "step": 2,
      "actual": 21,
      "predicted": 20.6667
    },
    {
      "step": 3,
      "actual": 22,
      "predicted": 22
    }
  ],
  "warnings": []
}

From an MCP client

catalog.call {
  "slug": "forecast-eval",
  "params": {
    "values": [
      10,
      12,
      13,
      12,
      15,
      16,
      18,
      19,
      21,
      22
    ],
    "testSize": 3,
    "method": "drift"
  }
}

On the hosted connector at https://agent402.tools/mcp, catalog.call runs forecast-eval free (rate-limited, no wallet). Local install: npx -y agent402-mcp.

Errors and behavior

• values, testSize and method are required. An input the tool rejects returns an HTTP 4xx whose body carries error, tool, expected, required and example, so the caller can correct it.
• A paid call that ends in any status of 400 or above is not charged over x402, MPP or a prepaid credits key: settlement is cancelled when the tool fails. The exception is a Tempo push credential, a transfer the buyer sent before the call: it settles before the tool runs, so if the tool then fails the payment is recorded as a refund owed to the paying wallet.
• Free tier: no outbound network call leaves the server for this tool, so proof-of-work (16 leading zero bits of sha256) pays for it.
• A GET or HEAD to /api/forecast-eval returns the same 402 quote, so the price can be read without a body.
• An Idempotency-Key header makes a retried paid call replay the first 200 instead of charging again (an answer larger than 1 MB is not replayed).

Paid call (JavaScript agent)

import { wrapFetchWithPayment } from "@x402/fetch";
import { x402Client } from "@x402/core/client";
import { registerExactEvmScheme } from "@x402/evm/exact/client";
import { privateKeyToAccount } from "viem/accounts";

const client = new x402Client();
client.setSpendControls?.(false); // keep your own spending ceiling in code
registerExactEvmScheme(client, { signer: privateKeyToAccount(KEY) });
const payFetch = wrapFetchWithPayment(fetch, client);

const res = await payFetch("https://agent402.tools/api/forecast-eval", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({
    "values": [
      10,
      12,
      13,
      12,
      15,
      16,
      18,
      19,
      21,
      22
    ],
    "testSize": 3,
    "method": "drift"
  }),
});

No wallet? Pay with compute

Fetch a challenge, solve the sha256 puzzle (16 leading zero bits, a fraction of a second of CPU), and resend with the X-Pow-Solution header:

import { createHash } from "node:crypto";
const lz = (b) => { let t = 0; for (const x of b) { if (!x) { t += 8; continue; } t += Math.clz32(x) - 24; break; } return t; };
const c = await (await fetch("https://agent402.tools/api/pow/challenge?slug=forecast-eval")).json();
let n = 0;
while (lz(createHash("sha256").update(c.challenge + ":" + n).digest()) < c.difficulty) n++;
await fetch("https://agent402.tools/api/forecast-eval", { method: "POST", headers: { "X-Pow-Solution": c.token + ":" + n, "Content-Type": "application/json" }, body: JSON.stringify({"values":[10,12,13,12,15,16,18,19,21,22],"testSize":3,"method":"drift"}) });

Part of these workflows

Forecast backtest (MAPE + RMSE) is one step in these 2 skill packs, each sold as a single call:

• Trend analysis - Take any numeric time series - a stock's daily close, a FRED macro indicator, a treasury yield history - and run it through the full quantitative workup: descriptives, moving averages, trend line, outliers, optional correlation against a benchmark, and a deterministic forecast forward with a 95% prediction interval. Everything an analyst writes a notebook for, in one chain of cheap calls.
• Forecasting bake-off - Don't guess which forecasting method to trust. Backtest all four (naive/drift, SES, Holt, Holt-Winters) on a real series, rank by out-of-sample RMSE, then forecast forward with the winner and its 95% prediction interval. Method selection without the hand-waving.

Related tools